From Subprime Loans to Subprime Growth? Evidence for the Euro Area

Transcription

1 9TH JACQUES POLAK ANNUAL RESEARCH CONFERENCE NOVEMBER 13-14, 2008 From Subprime Loans to Subprime Growth? Evidence for the Euro Area Martin Čihák and Petya Koeva Brooks International Monetary Fund Paper presented at the 9th Jacques Polak Annual Research Conference Hosted by the International Monetary Fund Washington, DC November 13-14, 2008 The views expressed in this paper are those of the author(s) only, and the presence of them, or of links to them, on the IMF webse does not imply that the IMF, s Executive Board, or s management endorses or shares the views expressed in the paper.

2 WP/08/xx From Subprime Loans to Subprime Growth? Evidence for the Euro Area Martin Čihák and Petya Koeva Brooks DRAFT

3 2008 International Monetary Fund WP/08/xx IMF Working Paper European Department From Subprime Loans to Subprime Growth? Evidence for the Euro Area Prepared by Martin Čihák and Petya Koeva Brooks 1 Authorized for distribution by [Luc Everaert] November 2008 Abstract This Working Paper should not be reported as representing the views of the IMF. The views expressed in this Working Paper are those of the author(s) and do not necessarily represent those of the IMF or IMF policy. Working Papers describe research in progress by the author(s) and are published to elic comments and to further debate. The global financial crisis has highlighted the potential of financial condions in influencing real economic activy. We examine the linkages between the financial and real sectors in the euro area, finding that (i) bank loan supply responds negatively to declines in bank soundness; (ii) a cutback in bank loan supply has a negative impact on economic activy; (iii) a posive shock to the corporate bond spread leads to a significant negative response of industrial output; and (iv) risk indicators for the banking, corporate, and public sectors show an improvement since , followed by a deterioration in These estimates imply that the currently estimated bank losses would subtract some 2 percentage points from the euro area output (but wh considerable uncertainty around the estimates). JEL Classification Numbers: E44, G12, G21 Keywords: Euro area, financial sector, real sector, linkages, contingent claims approach Authors Addresses: mcihak@imf.org; pkoeva@imf.org 1 We would like to thank, whout implicating, Jörg Decressin for guidance, Chanpheng Dara for research assistance, and Marie Donnay, Heiko Hesse, Luc Laeven, Kevin Ross, and participants in seminars at the IMF, the European Central Bank, and the European Commission for useful comments on earlier versions of the paper. All remaining errors are our own.

4 Contents Page I. Introduction...4 II. Empirical Evidence...5 A. Linkages Between Bank Characteristics and Lending Behavior...6 B. Linkages Between Bank Loan Supply and Aggregate Output...9 C. Linkages Between Corporate Financing Condions and Economic Activy...11 D. Risk Transfers Between Banks and Other Sectors: Contingent Claims Analysis..12 III. Quantative Implications...14 IV. Conclusions...16 References...35 Tables 1. Demand and Supply in the Disequilibrium Model, OLS Regression of Output on Loans First Stage IV Regression: Loans on Money Demand Shocks Second Stage IV Regression of Output on Loans...25 Figures 1. Euro Area: Money Market and Retail Lending Rates, Euro Area: Changes in Cred Standards to Enterprises and Households, Euro Area: Corporate and Equy Market Prices, Euro Area: Growth in Bank Loans and Securies Issuance, Euro Area: Distance to Default for Banks, Euro Area: Excess Demand for Loans, Euro Area: Growth in Real Output and Bank Loans, Euro Area: Corporate Debt Issuance, Euro Area: Response of Annual Growth in Industrial Production to One Standard Deviation Innovation in Corporate Spread Euro Area: Estimated Default Probabily (Banks and Non-Banks), Euro Area: Estimated Default Probabily (Public Sector), Capalization in Euro Area Banks, Appendixes I. Calculating the Distance to Default...26 II. Identifying the Linkage Between Bank Loan Supply and Aggregate Output...29 III. Contingent Claims Analysis: A Primer...32

5 I. INTRODUCTION The ongoing financial crisis has underscored the importance of financial linkages among countries, and the impact of financial condions on real economic activy (e.g., Strauss- Kahn, 2008). This paper studies empirically linkages between the financial and real sectors in the euro area. To assess the robustness of the main results, we use a battery of possible estimation approaches. Since mid-2007, the sub-prime mortgage crisis in the Uned States has sparked a reassessment of risk across global markets. Risk premia in money and cred markets have spiked, raising the cost of interbank and corporate financing, including in the euro area. The suation worsened substantially in September October 2008, when key money market indicators the 3-month spreads over policy rates, use of central bank facilies, and measures market segmentation on the basis of cred risk all rose to unprecedented levels. The impact of the global financial crisis on the euro-area real sector is an important, and still open, question. It may be too early to observe in full how the deterioration in financing condions will affect the euro-area economy. Nonetheless, is useful to examine the linkages between the financial and real sectors in the euro area, using a combination of past and recent data. The tight financial condions associated wh the crisis affect euro-area activy through four main channels, namely: First, the increase in bank funding costs (due to higher money market premia and rates) may be passed on to firms and consumers via higher lending rates. Indeed, of retail lending rates have gone up somewhat since mid-2007, even though this has been largely a continuation of a previous trend (Figure 1). Second, in response to their own deteriorated balance sheets and financial condions, banks may lim the amount of cred available to borrowers for any given price. This could be in the form of stricter lending standards. Bank lending surveys indicate a considerable tightening in quantative bank lending condions since mid-2007, suggesting that the cred cycle has turned (Figure 2). Third, the costs of corporate bond and equy financing may also be higher, liming the scope for substution from bank financing. The corporate bond and cred default spreads of all maturies and ratings have increased sharply, and the stock market has fallen since the start of the crisis (Figure 3). Fourth, tighter financing condions could create financial accelerator effects by depressing asset prices and reducing the value of collateral. Available data indeed confirm that asset prices have declined precipously (Figure 3); this has an impact on collateral values.

6 5 The recent data show that monetary aggregates have decelerated, as has bank cred to the private sector. This is a combination of a continued slowdown of bank cred to households and a starting slowdown in the (still relatively rapid) growth of corporate cred. Equy and bond issuance by non-financial firms has also decelerated (Figure 4). To preview our findings, we find that the financial tightening affects euro-area activy through several channels, including an increase in bank funding costs, bank cred rationing, increased costs of corporate bond and equy financing, and depressed collateral values. Based on a set of closely linked empirical approaches, we find that (i) bank loan supply responds negatively to declines in banks soundness; (ii) a cutback in bank loan supply is likely to have a negative impact on economic activy; (iii) a posive shock to the corporate bond spread leads to a significant negative response of industrial output; and (iv) risk indicators for the banking sector, the non-bank corporate sectors, and the public sector show a steady improvement since , followed by a deterioration in Combining the existing expert estimates of financial sector losses wh our econometric estimates of the relationship between financial sector losses and aggregate output, we conclude that the currently estimated banking sector losses would translate into a negative 2 percentage point impact on real output in the euro area (wh considerable uncertainty around this estimate). The remainder of the paper is structured as follows. Section II analyzes the empirical evidence on the financial-real linkages, using a battery of approaches. Section III puts the individual approaches together and quantifies the implications of the results. Section IV concludes. II. EMPIRICAL EVIDENCE We examine empirically the linkages between the financial and real sectors in several alternative but complementary ways. In the next four sub-sections, we focus on: A. Linkages between bank characteristics and lending behavior. This analysis helps us understand how financing condions for banks, which are a crucial part of the financial intermediation in Europe, translate into banks lending behavior, and thereby into financial condions of banks clients. B. Linkages between bank loan supply and aggregate output. This part of the analysis allows us to examine the relationship between the supply of bank cred and economic activy. Subsequently, we link the analysis in part B wh the analysis in part A, to examine the linkages between bank characteristics, bank lending, and aggregate output performance (the so-called bank lending channel ). C. Linkages between corporate sector financing condions and economic activy, using data on corporate bond spreads and output. This part of the analysis allows us to gauge how a change in corporate sector financing condions affects industrial output.

7 6 D. Risk transfers between banks, non-bank companies, and the public sector, using a combination of sectoral balance sheets and market-based data. A. Linkages Between Bank Characteristics and Lending Behavior Is bank loan supply in the euro area adversely affected by deteriorating financing condions? If this is correct, means that banks are not able to fully shield their loan portfolios from changes in financing costs. Most of the lerature on the bank lending channel deals wh the U.S. economy (e.g., Bernanke and Blinder, 1988; Bernanke and Gertler, 1995). It generally finds strong evidence that banks decrease their loan supply in response to tighter financing condions (in particular for small balance sheet-constrained banks), although there is ltle evidence that the cutback in bank loan supply leads to lower real activy (e.g., Ashcraft, 2003; Driscoll, 2003). The fact that banks still finance the bulk of investment in Europe constutes a good reason for investigating their impact on the monetary transmission process. However, the empirical evidence on the bank lending channel in Europe has been mixed. Several studies tested for the existence of a bank lending channel across euro-area countries (De Bondt, 1999; Favero, Giavazzi, and Flabbi, 1999; Altunbaş, Fazylov, and Molyneux, 2002; Angeloni and Ehrmann, 2003), and a number of studies examined the lending channel for individual countries (Angeloni and others (2003), for several countries; Kakes and Sturm (2002) for Germany, and Iacoviello and Minetti (2008) on four European housing markets). 2 The results from these studies are inconclusive, suggesting that the bank lending channel may be operating significantly in Germany, Italy and Greece, while appears not to be important in some other euro area countries. Most of these studies focus on the first (necessary) condion for the existence of a bank lending channel (i.e., that bank loan supply is affected by higher financing costs), whout examining whether the decline in cred supply has an adverse effect on the real economy. Identifying the determinants of cred developments is complicated by the interplay of cyclical and long-term factors that influence both cred demand and cred supply. On the cred demand side, these include a combination of cyclical developments and structural shifts. On the cred supply side, the impact of the economic downturn on financial markets and the financial suation of the banks seems to have influenced their lending. Moreover, banks in the euro area have gone through important structural changes that included a move 2 In addion, there are also other studies covering banking intermediation in the euro area, but focus on other issues than the bank lending channel. For instance, Bruggeman and Donnay (2003) estimate a monthly monetary model wh banking intermediation for the euro area in , but instead of the bank lending channel focus on the relationship between short-term market interest rates, retail interest rates, and inflation.

8 7 from relationship-based banking to more market-based banking, and a growing role of securization (e.g., Gambacorta and others, 2008). A rough tool for distinguishing cred supply and demand factors are the bank lending surveys. These surveys, organized by the Eurosystem central banks since 2003, summarize responses of senior loan officers regarding loan demand and changes in their bank s lending policy in the previous quarter. The changes in demand condions and cred standards in a preceding quarter are summarized by a difference between posive and negative responses, in percent of all responses ( net percentage ). 3 When interpreting the results of the survey, one needs to take into account the qualative, subjective nature of the survey data. In particular, experience from similar surveys suggests that bankers responses may be biased towards tightening, and therefore a zero net percentage may in fact mean a slight easing. Wh that in mind, the latest survey data indicate that bank cred standards have tightened considerably since mid-2007, both for households and for enterprises (Figure 2). The three most important factors listed by banks when explaining changes in cred standards were those related to the perception of risk. Empirically, there is some basic evidence that the bank lending surveys contain useful information about subsequent macroeconomic developments. For example, there is a posive correlation between the quarter-to-quarter growth of real GDP and lagged values of the net percentage balance of loan demand (interestingly, the correlation coefficient is the same, 0.41, for both household lending and enterprise lending); and there is a negative correlation between the quarter-to-quarter growth of real GDP and the lagged net percentage balance reflecting cred standards (the correlation coefficient being for household lending and for enterprise lending). This suggests that both the loan demand and the lending standards are procyclical. The time series of lending surveys are too short to allow for a more elaborate analysis or to test for breaks in the correlations. To analyze the bank cred channel in the euro area, we use a supply-demand disequilibrium model. Equilibrium approaches, such as VEC/VAR models or single-equation estimates can provide only a limed answer to the causes of cred slowdown, because they do not address the question whether the demand or supply function determines the cred. Following the examples of Pazarbasioglu (1997), and Barajas and Steiner (2002), a cred demand- and a cred supply-function are estimated under the restriction that the minimum of the two determines the cred. This strategy avoids the identification problem of equilibrium models, and allows to make a statement on the existence of a cred crunch. 3 A posive net balance on demand side means more demand, while on supply side means less supply. See Berg and others (2005) for an overview of the methodology of the surveys.

9 8 The disequilibrium model is estimated wh bank-by-bank panel data for 50 largest euro-area banks from 1997 to The specification of the demand side follows Bundesbank (2002). The specification of the supply side is close to Pazarbasioglu (1997), but wh the distance to default (DD) among the supply-side variables. The distance to default was used to approximate banking sector vulnerabily as a possible source of cred supply strain (see Appendix I for details). The DD for this estimate was calculated for each individual bank. The advantage of using individual bank data is that allows for testing whether weaker banks are more likely to restrain their cred. Nonetheless, we have also calculated the aggregate DD for a portfolio of euro area banks (using a methodology explained in Appendix I and also in De Nicolò and others (2005) and Čihák (2007)). To provide an illustration of the overall developments in the DD, Figure 5 shows the development of the portfolio DD for daily data since early 1990s. The portfolio DD has generally been above 2, except for a brief period in 2003 (which can be linked to weakness in German banks), and except for the most recent period: in October 2008, the portfolio DD reached zero, s lowest recorded value. The estimated model provides a plausible explanation of the factors contributing to cred developments in the major euro area banks (Table 1). All the key coefficients have the expected signs and are significant. The model explains year-on-year real growth rates of customer loans as a function of a bank s distance to default (wh an expected posive sign, as higher distance to default is associated wh greater soundness, making easier for banks to expand lending), real GDP growth rate as a proxy for overall economic activy (posive sign), lending rate and net interest margin (expected negative signs, reflecting more expensive lending for borrowers), and bank size approximated by total value of loans (expected negative sign). The key variable of interest is the distance to default, which captures the effect of bank financial condions on cred supply. Based on the estimated coefficients, the effect of bank soundness on loan supply is significant, but relatively small. For example, the estimate implies that a one-standarddeviation drop in the distance to default is associated wh a year-on-year real growth of cred that is 1.5 percentage points lower than otherwise. Figure 6 illustrates the development of the excess demand for cred in the model. It is an aggregate number, calculated by aggregating the demand and supply estimates for all the individual banks. The figure suggests that in 2000 there was a period of excess supply of cred, while 2003 and 2004 were characterized by excess demand for cred. Since then, demand and supply have been relatively balanced. 4 Data are from the BankScope database by Bureau van Dijk for To explain the factors contributing to cred developments, the following variables are used: total bank assets, total loans, shareholders equy, short-term liabilies, long-term liabilies, liquid holdings (cash, ECB and other financial instutions securies, and government securies), equy price data ( last price, daily), and equy shares outstanding (daily).

10 9 As a robustness test, we ran the same model, replacing the distance to default by the probabily of distress (PD) calculated in Poghosyan and Čihák (2008). The PD has the expected (negative) sign in the loan supply (higher PD, i.e. lower soundness, implies lower supply of loans), and the coefficients of the other variables are largely unaffected. As another robustness check, we have performed this analysis also at the level of countries rather than the level of individual banks. Specifically, this means that instead of individual bank DDs, we have used the aggregated DDs for portfolios of banks in the individual Euro area countries (see Appendix I). This reduces the number of available observations, but allows for an easier link to the subsequent analysis (of linkages to aggregate output), which is also carried out at the level of countries rather than banks. In addion to the disequilibrium model presented in Table 1, a series of pairwise Granger causaly tests were run to assess the relationships between real cred growth, real output growth, and banking sector vulnerabily (approximated again by distance to default). The results of the exercise suggest that banking sector vulnerabily, measured by distance to default, is influenced by real GDP and real cred in the horizon of 2 4 quarters. The distance to default influences real cred, but not GDP, wh a lag of 6 quarters (detailed results available upon request). B. Linkages Between Bank Loan Supply and Aggregate Output Our next step is to examine the relationship between the supply of bank cred and economic activy. Declining loan supply may suppress economic activy if firms and households cannot replace completely the missing loans wh other funding. For this to hold, a substantial group of borrowers (firms or households) must be unable to insulate their spending from the reduction in bank cred. As the data illustrate, bank cred to the private sector and output do move together (Figure 7). But this does not necessarily mean that the supply of bank loans has a significant effect on output. An alternative (and equally plausible) possibily is that as economic activy slows, the demand for bank loans declines, leading to a posive relationship between the two series. Disentangling the demand and supply effects (i.e., solving the identification problem) is very hard, since these effects tend to occur at the same time but only the equilibrium outcome is observed. The identification problem can be addressed by using an instrumental variables (IV) technique to isolate the loan supply effect on real output. We use shocks to country-specific money demand as an instrument for shocks to loan supply, as first proposed by Driscoll (2004) in addressing a similar question for the Uned States. The logic behind this approach is based on the premise that country-specific shocks to money demand should lead to country-specific changes in the supply of loans, and therefore changes in output. This would allow to isolate the effect of loan supply on real activy.

11 10 The identification scheme involves the following three steps, wh all variables used in the regressions constructed as deviations from their cross-sectional mean values, as implied by the identification scheme (see Appendix II for details): The overall effect of bank cred on output is investigated by regressing output growth on the growth rate of bank loans (and s lagged value), as well as s own lagged values. The resulting coefficient will reflect both the supply and demand effects of bank cred on real activy. The shocks to money demand are recovered after estimating money demand functions for each euro-area country in the sample. Then the growth rate of bank loans is regressed on s lagged values and the estimated money demand shocks, in order to establish whether the latter are a good instrument for shocks to loan supply. The effect of bank cred on output (see first bullet) is re-estimated using the countryspecific shocks to money demand as instruments. The resulting coefficient of bank loans is indicative of the supply effect, as the demand effect has been stripped out. 5 The estimations are done using country-level data from 1999Q1 to 2008Q2. The sample includes 11 euro-area countries (Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, Netherlands, Portugal, Spain). 6 The key variables used in the analysis are: real GDP, M3, depos rates, and bank loans to non-financial corporations. For each country, the money supply (M3) and bank loan variables are deflated by the corresponding GDP deflator. Except for depos rates, all other variables are in logarhmic form. One issue when doing any analysis for euro area banks involving interest rate statistics is that harmonized data on interest rates are available from the ECB only from January For the earlier period, we have to rely on non-harmonized country-level data, available since January To address this issue, we carry out a robustness check by performing the same analysis only for the sub-period for which harmonized depos rate data are available, i.e. for the period since 2003Q1. The estimation results from the first step confirm the posive relationship between bank cred and economic activy. As shown in Table 2, real bank cred has a significant and posive effect on output. The size of the coefficient suggests that an increase in bank cred (in real terms) by 10 percentage points is associated wh an increase in real GDP by about 1.5 percentage points. 5 Assuming that shocks to loan demand and supply are posively correlated, we would expect the instrumented coefficient of bank loans to be smaller than the non-instrumented one. 6 Cyprus, Malta, Luxembourg, and Slovenia are not included due to data limations.

12 11 Turning to the second step, we find that posive money demand shocks are associated wh higher growth in bank loans. The shocks to money demand are constructed using estimates of country-specific money demand functions (see Appendix II). Their impact on bank loans is illustrated by the posive and significant coefficient of the (country-specific) residuals from the estimated money demand functions on the growth of bank loans, even after controlling for lagged values of output (Table 3). Therefore, the money demand shocks can be used as an instrument for loan demand in the next step. Once demand effects are taken into account, the loan supply effect on output is posive, and statistically significant, even though relatively small. As the results in Table 4 show, the coefficient of the bank loan variable is still posive but somewhat smaller than in the first step (0.10 instead of 0.15) when the instrumental variables estimation is implemented. Overall, the estimation results suggest that an increase in the supply of bank loans by 10 percentage points is likely to lead to an increase in real GDP by about 1 percentage point. Therefore, our analysis implies that a cutback in bank loan supply is likely to have a negative impact on economic activy. As another robustness check, we have tried an alternative approach in which we have introduced the difference between unsecured and government-backed depos rates as an addional instrument for cred risk. This is motivated by the approach of Greenlaw and others (2008), who used the Treasury-Eurodollar (TED) spread as an instrument for cred supply in the Uned States. A weakness of the TED spread is that may be influenced by flight to qualy flows that move Treasury bill yields, as well as the funding pressures that drive LIBOR rates. Nonetheless, as the difference between unsecured and governmentbacked depos rates, the TED spread could potentially provide a measure of cred risk, which is likely to be correlated wh cred supply. Using a similar line of argumentation, we have introduced in our regressions the spread between the 3-month euro LIBOR rate and the German government bund rate, to instrument for cred risk. This robustness check has not affected the quantative predictions from the main regression. C. Linkages Between Corporate Financing Condions and Economic Activy Turning to the question of how corporate sector financing condions affect activy, we analyze the relationship between the corporate bond spread and the euro area output. The corporate bond spread is defined as the difference between the yield on a corporate bond (risky asset) of a given matury and qualy and the yield on a government bond (riskless asset) of the same matury. The corporate bond risk premium has been shown to be a good predictor of real activy in the Uned States (Chan-Lau and Ivaschenko, 2002; Mody and Taylor, 2004) and there were some early results suggesting a similar relationship for the euro area (De Bondt, 2002). There are a number of reasons why the corporate bond spread can be a good predictor of real activy. First, financial instruments, such as corporate bonds, ultimately represent claims on

13 12 the real economy. Financial information is readily available at high frequencies and transmted relatively rapidly compared wh economic information such as that on output. Therefore, financial prices such as corporate bond spreads could provide useful leading information on economic activy. Second, corporate-sovereign bond spreads are a key measure of the cred terms. Their role in predicting output is consistent wh the presence of a financial accelerator in the economy, i.e. wh the presence of a mechanism linking the condion of borrower balance sheets to the terms of cred, and hence to the demand for capal. Third, the bond market has become a relevant source of corporate financing in the euro area. Since 1999, the euro-area market for corporate debt securies has grown tremendously (Figure 8). Fourth, as corporate bond spreads tend to move together wh the tightness of bank lending standards (for evidence in the Uned States, see Duca, 1999; Gertler and Lown, 2000), they can also be treated as a proxy for corporate sector financing condions. At the euro-area level, aggregate data on corporate bond yields are available for securies of different maturies and qualy. The spreads for AAA, AA, A, and BBB 7-year corporate bonds in the euro area (in relation to a 7-year government bond) are shown in Figure 3. Given the high frequency nature of these data, we use monthly industrial production (rather than real GDP) as an indicator for economic activy. The analysis is conducted using vector autoregression (VAR) estimates run over the period from 1999M1 to 2008M1. The key variables in the regressions are the corporate bond spread, the annual growth in industrial production, and the annual change in the real effective exchange rate. Our baseline specification of the VAR includes three lags; as a robustness check, we also experiment wh increasing the number of lags in the VAR. The corporate bond spread is defined as BBB yield minus government bond yield in the benchmark regressions; as a sensivy analysis, we also conduct the same analysis for AA and A rated bonds. The estimation results show that a posive shock to the corporate bond spread leads to a significant negative response of industrial output. The impulse responses of the baseline regressions are shown in Figure 9. The results illustrate that a one-standard-deviation shock to the corporate bond yield (about 60 basis points) has an adverse effect on the growth rate of industrial output, which peaks at about 0.25 percent in 8 20 months. This effect is statistically significant, as shown by the 95 percent confidence bands. A limation of these estimates is that simultaney might be an issue in the basic VAR estimation (we are not using a structural VAR). Nonetheless, these results are fairly robust across alternative specifications. D. Risk Transfers Between Banks and Other Sectors: Contingent Claims Analysis How are risks transmted between the corporate sector, the financial sector, and the public sector in the euro area? One way of addressing this question is to employ a contingent claims

14 13 analysis (CCA). The CCA is an improved version of the balance-sheet approach, which incorporates not only accounting data, but also information contained in market prices. It starts wh information on the size and structure of assets and liabilies of key economic sectors, wh the aim of assessing the extent of currency and matury mismatches, or imbalances in the debt and equy structure. Given that economy-wide balance sheet data do not provide a full picture of all the risks (because of the contingent nature of many risks), and given that they are usually valued at book value, they do not capture changes in the likelihood of default related to recent market developments. To provide a more complete picture of the risks inherent in a balance sheet, the CCA values assets using marked-tomarket prices and incorporates contingent liabilies. To the best of our knowledge, this is the first time in the lerature that CCA is used to identify vulnerabilies in the corporate, banking, and public sectors in the euro area (and to estimate the associated value of risk transfer across the balance sheets). Appendix III provides more details on the CCA methodology employed in this paper. Default indicators for both the banking system and non-bank corporate sector show an improvement from the lows of , followed a major deterioration in the second half of 2007 and 2008 (Figure 10). The distribution of default risk by assets confirms the general improvement in both banking system and non-bank corporate sector indicators, wh the riskiest banks (those wh the highest default probabily) accounting for a smaller percentage of total assets over time. Expected losses for the banking system have declined steadily since ; the indicator has deteriorated in the second half 2007 and The generally posive trend of the last 5 years reflected rising equy valuations and declining volatilies, as balance-sheet structures have improved and non-performing assets have declined (the measure of expected losses for the largest banks moves closely wh the overall NPL ratio and tends to lead changes in the NPL ratio by 1 2 quarters). For the public sector balance sheets, Figure 11 shows an improvement in the soundness over the past five years. It shows the estimated default probabily when 100 percent of expected losses of the banks are assumed to be guaranteed by the sovereign (solid line) and when expected losses are excluded (dashed line). The figure suggests a gradual decline in default probabilies since For the later period for which daily information is available, the sovereign spread and default probabily move in line wh the downward trend in spreads on government debt quoted by the market. The estimated probabily of default for the public sector is substantially lower than those reported in previous CCA studies. This reflects the fact that previous applications of CCA (surveyed in Appendix III) have covered emerging market economies (e.g., Brazil, Turkey, Thailand, Indonesia), while this analysis focuses on an advanced economy (or, more specifically, the set of advanced economies that form the euro area). For example, Gray and Jones (2006) examine the 1-year sovereign default probabily in Indonesia in , and find that most of the period was in the range of 2 6 percent (wh a spike to 11 percent

15 14 in 2001). The other studies surveyed in Appendix III find numbers in a similar range. The numbers reported for euro area in Figure 11 are lower by an order of magnude, being generally below 0.2 percent. This is consistent wh the high sovereign ratings of euro area countries. The financial turbulence experienced since mid-2007 has led to increased volatily in a variety of risk indicators. Figure 10 shows the development in the estimated probabilies of default for the large banks and non-bank corporations in the euro area. The two move broadly in line, but banks have so far been affected much more by the recent financial turmoil. The global market turmoil experienced since mid-2007 caused a substantial worsening in risk indicators for the banking system. A combination of reduced market capalization and increase in s volatily decreased implied assets and increased their volatily, leading to a decline in distance-to-default measures and increases in expected losses. These developments reflect the increased market volatily, in combination wh declining capalization and lower earnings. Figure 11 shows the estimated public sector default probabilies, illustrating the impact of the financial sector instabily on public sector soundness. The public sector probabily of default is, but assuming that the public sector would be willing to guarantee the large banks (indicated by the line wh guarantees ), the impact on public sector stabily would be considerable. III. QUANTITATIVE IMPLICATIONS What do the calculations imply quantatively for euro-area developments? Based on the estimates presented in the preceding section, we can calculate the potential impact of banking sector losses on future economic growth in the euro area. Specifically, the current estimates of losses in the banking sector would imply a negative 2 percentage point impact on GDP in the euro area. Here is how this estimate is derived: A natural starting point are estimated losses in euro-area commercial banks. These losses have been somewhat of a moving target, as the crisis evolved from the subprime crisis in the Uned States into a global crisis. The estimated sub-prime related losses in euro-area banks as of March 2008 were only US$45, as reported in IMF s April 2008 Global Financial Stabily Report (IMF, 2008a); the latest estimates of the total exceptional losses in euro-area global banks (which combines the sub-prime related losses wh the exceptional part of losses generated on European assets) may be as large as 10 times that amount. The estimated losses for the whole of Europe

16 15 were even larger, but substantial chunks of these losses were in global banks based in the Uned Kingdom and Swzerland. 7 These estimated losses correspond to about 14 percent of the euro-area banks capal and reserves. If nothing else happened, the ratio of capal to total (unweighted) assets in euro area banks, currently at 5.6 percent (Figure 12), would decline to 4.8 percent, and the banks leverage would increase correspondingly. One way to think about the potential impact of these losses on asset growth is to ask how much would assets have to shrink to prevent the leverage ratio from declining. Keeping leverage ratio at 5.6 percent would at the new (decreased) level of capal require that assets go down by 14 percent. The impact of bank losses on lending (and thereby on output), can be larger if banks (or their regulators) aim to de-leverage, i.e. decrease their leverage target, which is que likely given the overall increase in risk aversion (see, e.g., IMF, 2008), and if they get h by addional shocks, such as stock price declines. 8 To increase leverage ratio to 5.9 percent (the sample maximum in Figure 12), assets would have to go down by 19 percent. Just to illustrate the sensivy of this result, increasing the leverage to 7 percent (which is beyond the recent historical experience, but not implausible) would in this suation imply a decline in assets by 31 percent. From the estimate in Section II.B, a decline in the supply of bank loans by 10 percentage points is likely to lead to a decline in real GDP by about 1 percentage point. Following up on the calculations from the previous bullet point, a loan decline by 14 percent therefore corresponds to 1.4 percentage points negative impact on real GDP; a loan decline by 19 percent corresponds to 1.9 percentage points negative impact on real GDP; and a loan decline by 31 percent corresponds to 3.1 percentage points negative impact on real GDP. An alternative approach to analyzing the recent developments is to start from the recent changes in distance to default and their estimated impact on loan supply. As previously discussed, distance to default is a market-based indicator that incorporates market participants view on banks suation and outlook. It can therefore provide an alternative assessment of the likely impact of the shocks that h the banks. 7 The calculations underlying the October 2008 Global Financial Stabily Report (IMF, 2008b) suggest total exceptional losses in large banks in continental Europe close to $500 billion. 8 On the other hand, these effects can be migated to some extent if banks achieve the increases in their capalto-asset ratios (decrease leverage) through capal injections rather than (or in addion to) asset manipulation.

17 16 To calculate the impact on banks lending, we can use the results of the distance-to-default calculations in Section II.A. The average distance to default was 0.0 in October 2008 (Figure 5), compared to 8.0 in July Using the estimates in Section II.A, this translates into a decline in real cred by 19 percentage points. That in turn (using the estimates from Section II.B.) translates into a real GDP decline by some 1.9 percentage points. In other words, this method yields a broadly similar estimate of the likely GDP impact than the method based on projected capal losses. The difference between the two approaches reflects a variety of factors. This includes the extent to which the banks will (or will not) be recapalized. The extent of recapalization is not trivial to estimate, making the market s guess a useful alternative input. The above calculations illustrate that there are linkages between the financial sector soundness and real economic developments. They also illustrate the challenges of quantifying the exact relationship, and the uncertainties surrounding the estimates. We find that based on current information, the likely impact of the recent and projected banking losses on output around 2 percentage points (wh substantial uncertainty relating to the impact of the recapalization and more generally to the impact on market confidence). IV. CONCLUSIONS This paper examines the impact of financing condions on real economic activy in the euro area, exploring some key linkages between the financial and real sectors. To explore the evidence, applies a broad range of empirical approaches and estimation methods to banklevel, country-level and aggregate data. The main findings are as follows: First, a deterioration in the financial health of banks could translate into lower bank loan supply; this effect is statistically significant, but quantatively small. Second, a cutback in bank loan supply is likely to have a negative impact on economic activy in the euro area; again, this effect is statistically significant, but relatively small. These findings are not dissimilar to the lerature on the bank lending channel in the Uned States, which generally finds strong evidence banks decrease their loan supply in response to tighter financing condions, but ltle evidence that the cutback in bank loan supply leads to lower real activy. Third, higher costs of corporate bond financing (which could also reflect broader financial condions in the economy) tend to lead to a significant negative response of industrial production growth. Last but not least, risk indicators for the banking, corporate, and public sectors in the euro area (derived from the CCA) show a steady improvement in balance sheets since

18 , followed by a major deterioration in 2007 and especially 2008, reflecting a combination of the increased market volatily and lower capalization. Condions as of October 2008 were the worst in the whole sample period (since early 1990s). The estimates presented in this paper can be used to calculate the potential impact of the banking sector losses on future economic growth in the euro area. They suggest that current estimates of losses in the banking sector would mean a negative 2 percentage point impact on GDP in the euro area, but wh substantial uncertainty around this estimate.

19 Figure 1. Euro Area: Money Market and Retail Lending Rates 5.5 Money Market Interest Rates (percentages per annum) Overnight interest rate (EONIA) 3-month EONIA swap 3-month EURIBOR /01/07 02/26/07 04/23/07 06/18/07 08/13/07 10/08/07 12/03/07 01/28/08 03/24/08 05/19/08 07/14/08 09/08/08 10 Retail Lending Rates to Non-financial Corporations and Households (percentages per annum, rates on new business) Loans to non-financial corporations over EUR 1 million at floating rate and up to 1 year inial rate fixation Loans to non-financial corporations up to EUR 1 million at floating rate and up to 1 year inial rate fixation Loans to households for consumption over 1 and up to 5 years inial rate fixation 2 Loans to households for house purchase at floating rate and up to 1 year inial rate fixation 0 0 Jan-03 Jul-03 Jan-04 Jul-04 Jan-05 Jul-05 Jan-06 Jul-06 Jan-07 Jul-07 Jan-08 Jul-08 Sources: Deutsche Bundesbank, Datastream.

20 19 Figure 2. Euro Area: Changes in Cred Standards to Enterprises and Households, Changes in cred standards applied to the approval of loans to or cred lines to enterprises (net percentages of banks reporting tightening cred standards) Competion from other banks Realized Expected Costs related to bank's capal Expectations regarding general economic activy Industry or firmspecific outlook Q1 2003Q3 2004Q1 2004Q3 2005Q1 2005Q3 2006Q1 2006Q3 2007Q1 2007Q3 2008Q1 2005Q4 2006Q2 2006Q4 2007Q2 2007Q4 2008Q2 2006Q1 2006Q3 2007Q1 2007Q3 2008Q1 2005Q4 2006Q2 2006Q4 2007Q2 2007Q4 2008Q2 2006Q1 2006Q3 2007Q1 2007Q3 2008Q1 2008Q Changes in cred standards applied to the approval of loans to households for house purchase (net percentages of banks reporting tightening cred standards) Realized Expected Competion from other banks Housing market prospects Expectations regarding general economic activy Cost of funds and balance sheet constraints Q1 2003Q3 2004Q1 2004Q3 2005Q1 2005Q3 2006Q1 2006Q3 2007Q1 2007Q3 2008Q1 2005Q4 2006Q2 2006Q4 2007Q2 2007Q4 2008Q2 2006Q1 2006Q3 2007Q1 2007Q3 2008Q1 2005Q4 2006Q2 2006Q4 2007Q2 2007Q4 2008Q2 2006Q1 2006Q3 2007Q1 2007Q3 2008Q1 2008Q3 Source: European Central Bank.

21 Figure 3. Euro Area: Corporate and Equy Market Prices, (Yields in percent, spreads in basis points) Selected IBoxx Yields IBOXX Euro Corporates A Rated All Maturies IBOXX Euro Corporates AA Rated All Maturies IBOXX Euro Corporates AAA Rated All Maturies IBOXX Euro Corporates All Maturies IBOXX Euro Corporates BBB Rated All Maturies /1/07 2/1/07 3/1/07 4/1/07 5/1/07 6/1/07 7/1/07 8/1/07 9/1/07 10/1/07 11/1/07 12/1/07 1/1/08 2/1/08 3/1/08 4/1/08 5/1/08 6/1/08 7/1/08 8/1/08 9/1/08 10/1/08 11/1/ A rated; all maturies AA rated, all maturies AAA rated; all maturies All maturies BBB rated; all maturies Selected IBoxx Spreads 1/ /1/07 2/1/07 3/1/07 4/1/07 5/1/07 6/1/07 7/1/07 8/1/07 9/1/07 10/1/07 11/1/07 12/1/07 1/1/08 2/1/08 3/1/08 4/1/08 5/1/08 6/1/08 7/1/08 8/1/08 9/1/08 10/1/08 11/1/ Euro Stoxx 50, Index, Jan. 1, 2007=100 Impl. volatily (in percent, right scale) Stock prices /1/07 2/1/07 3/1/07 4/1/07 5/1/07 6/1/07 7/1/07 8/1/07 9/1/07 10/1/07 11/1/07 12/1/07 1/1/08 2/1/08 3/1/08 4/1/08 5/1/08 6/1/08 7/1/08 8/1/08 9/1/08 10/1/08 11/1/08

22 Figure 4. Euro Area: Growth in Bank Loans and Securies Issuance, (Percent) Bank Loans to Non-financial Corporations and Households (annual growth rates) Loans to households Loans to non-financial corporations Jan-03 Jul-03 Jan-04 Jul-04 Jan-05 Jul-05 Jan-06 Jul-06 Jan-07 Jul-07 Jan-08 Jul Debt Securies Issuance (annual growth rates) Non-monetary financial instutions Non-financial corporations MFIs Jan-03 Jul-03 Jan-04 Jul-04 Jan-05 Jul-05 Jan-06 Jul-06 Jan-07 Jul-07 Jan-08 Jul Equies Issuance (annual growth rates) Non-monetary financial instutions Non-financial corporations MFIs Jan-03 Jul-03 Jan-04 Jul-04 Jan-05 Jul-05 Jan-06 Jul-06 Jan-07 Jul-07 Jan-08 Jul-08 Source: ECB.

23 22 Figure 5. Euro Area: Distance to Default for Banks, (daily data; higher values mean higher soundness) Distance to default, portfolio of euro area banks /1/1991 1/1/1992 1/1/1993 1/1/1994 1/1/1995 1/1/1996 1/1/1997 1/1/1998 1/1/1999 1/1/2000 1/1/2001 1/1/2002 1/1/2003 1/1/2004 1/1/2005 1/1/2006 1/1/2007 1/1/2008 Figure 6. Euro Area: Excess Demand for Loans, % ofcred supply excess demand for cred demand minus supply +2 st.dev. -2 st.dev excess supply of cred Figure 7. Euro Area: Growth in Real Output and Bank Loans, Real year-on-year change (%) Bank lending to private sector 0 Mar-00 Sep-00 Mar-01 Sep-01 Mar-02 Sep-02 Mar-03 Sep-03 Mar-04 Sep-04 Mar-05 Sep-05 Mar-06 Sep-06 Mar-07 Sep-07 Mar-08 Sep-08 Gross domestic product

24 23 Figure 8. Euro Area: Corporate Debt Issuance, Debt Securies (amount outstanding, EUR billions) Non-financial corporations Non-monetary financial instutions MFIs Jan-90 Jan-92 Jan-94 Jan-96 Jan-98 Jan-00 Jan-02 Jan-04 Jan-06 0 Figure 9. Euro Area: Response of Annual Growth in Industrial Production to One Standard Deviation Innovation in Corporate Bond Spread Figure 10. Euro Area: Estimated Default Probabily (Banks and Non-Banks), Probabily of default (next 12 months) non-bank corporations 1/1/91 1/1/92 1/1/93 1/1/94 1/1/95 1/1/96 1/1/97 1/1/98 1/1/99 1/1/00 1/1/01 1/1/02 1/1/03 1/1/04 1/1/05 1/1/06 1/1/07 1/1/08 banks

25 24 Figure 11. Euro Area: Estimated Default Probabily (Public Sector), /1/97 7/1/97 1/1/98 7/1/98 1/1/99 7/1/99 1/1/00 7/1/00 1/1/01 7/1/01 1/1/02 7/1/02 1/1/03 7/1/03 1/1/04 7/1/04 1/1/05 7/1/05 1/1/06 7/1/06 1/1/07 7/1/07 1/1/08 7/1/08 Probabily of default (next 12 months) whout guarantees wh guarantees 6.0 Figure 12. Capalization in Euro Area Banks, (Capal and Reserves as percentage of Total Assets) Capal and reserves to total assets (percent) Sep-97 Sep-98 Sep-99 Sep-00 Sep-01 Sep-02 Sep-03 Sep-04 Sep-05 Sep-06 Sep-07 Sep-08 Table 1. Demand and Supply in the Disequilibrium Model, / (Dependent variable: year-on-year real growth rate of a bank s total cred) Explanatory variables Demand Supply Parameter Std. Error Parameter Std. Error Constant Real GDP growth Lending rate Net interest margin Distance to default Log (total loans) Source: Authors calculations based on data from BankScope and DataStream. 1/ Maximum likelihood estimation. Log likelihood =